Don't mark the region tracker page as allocated #899
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mconst
force-pushed
the
region_tracker_alloc
branch
2 times, most recently
from
ffb8f6d to
b798a4a
Compare
|
Okay, I think this is ready. Sorry for the delay! The first two commits are fixes for preexisting region-tracker bugs I ran into; the last one is the main change. |
|
Ah nice, ya I like this! I think this conflicted with your other PR that I just merged. Can you rebase and I'll merge this one too? |
cberner
approved these changes
Nov 25, 2024
If we relocate the region tracker page during compaction and then the transaction aborts, the allocation of the new tracker page gets rolled back, but the free of the old tracker page doesn't! This leaves us in an inconsistent state with no tracker page allocated. So we need to use allocate_non_transactional() here, which guarantees the page will remain allocated even if the rest of the transaction rolls back.
If repair fails, our allocator state is inconsistent; don't try to write it to disk
When writing the allocator state table, it's better not to include the region tracker page as one of the allocated pages. Instead, we should allocate the region tracker page during quick-repair after loading the rest of the allocator state, exactly like we do for full repair. This avoids an unlikely corruption that could happen for databases larger than 4 TB, if we crash on shutdown after writing out a new region tracker page number but before clearing the recovery_required bit. It also means that in the future, when we drop support for the old allocator state format, we'll be able to get rid of the region tracker page entirely instead of having to keep around some of the allocation code for compatibility.
mconst
force-pushed
the
region_tracker_alloc
branch
from
b798a4a to
420791e
Compare
|
Oh yeah, sorry about that! Here's the rebased version. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When writing the allocator state table, it's better not to include the region tracker page as one of the allocated pages. Instead, we should allocate the region tracker page during quick-repair after loading the rest of the allocator state, exactly like we do for full repair.
This avoids an unlikely corruption that could happen for databases larger than 4 TB, if we crash on shutdown after writing out a new region tracker page number but before clearing the recovery_required bit. It also means that in the future, when we drop support for the old allocator state format, we'll be able to get rid of the region tracker page entirely instead of having to keep around some of the allocation code for compatibility.